使用Mysql EXPLAIN分析、优化SQL语句 （续）

今天优化了多条SQL语句。都是EXPLAIN的功劳，分析SQL子句的执行顺序和执行情况，一木了然，下来看具体分析：

［优化多表联合查询］

explain SELECT sql_no_cache pker.*,pk.*  FROM ng_game_pk AS pk ,ng_game_pker AS pker  where pker.pkid = pk.id and (pker.act_uid = 1 OR pker.def_uid = 1) AND pk.type <>4  GROUP BY pk.id limit 10;
+----+-------------+-------+--------+----------------------------------+---------+---------+--------------------+--------+----------------------------------------------+
| id | select_type | table | type   | possible_keys                    | key     | key_len | ref                | rows   | Extra                                        |
+----+-------------+-------+--------+----------------------------------+---------+---------+--------------------+--------+----------------------------------------------+
|  1 | SIMPLE      | pker  | ALL    | pkid,act_def                     | NULL    | NULL    | NULL               | 177543 | Using where; Using temporary; Using filesort |
|  1 | SIMPLE      | pk    | eq_ref | PRIMARY,type,idx_type_status_uid | PRIMARY | 4       | pwgbk8.7.pker.pkid |      1 | Using where                                  |
+----+-------------+-------+--------+----------------------------------+---------+---------+--------------------+--------+----------------------------------------------+

key:NULL,,,Extra:Using temporary,Using filesort都会造成语句很慢。type:ALL 全表扫描，没有比这个更糟糕的了。

下面作出改动：

explain SELECT sql_no_cache pker.*,pk.*  FROM ng_game_pk AS pk ,ng_game_pker AS pker  where pker.pkid = pk.id and (pker.act_uid = 1 OR pker.def_uid = 1) AND pk.type <>4  GROUP BY pker.pkid limit 10;
+----+-------------+-------+--------+----------------------------------+---------+---------+--------------------+------+-------------+
| id | select_type | table | type   | possible_keys                    | key     | key_len | ref                | rows | Extra       |
+----+-------------+-------+--------+----------------------------------+---------+---------+--------------------+------+-------------+
|  1 | SIMPLE      | pker  | index  | pkid,act_def                     | pkid    | 4       | NULL               |   10 | Using where |
|  1 | SIMPLE      | pk    | eq_ref | PRIMARY,type,idx_type_status_uid | PRIMARY | 4       | pwgbk8.7.pker.pkid |    1 | Using where |
+----+-------------+-------+--------+----------------------------------+---------+---------+--------------------+------+-------------+

比较下执行：

mysql> show profile for query 147;
+----------------------+----------+
| Status               | Duration |
+----------------------+----------+
| Creating tmp table   | 0.000372 |
| Copying to tmp table | 0.424248 |
| removing tmp table   | 0.002125 |
+----------------------+----------+

这里忽略了其他的执行。看到拷贝到临时表占用了大部分时间。这里需要解释下：Copying to tmp table是拷贝到内存，如果是Copying to tmp table on disk意味着内存空间不足，MySQL将会写临时表到磁盘。这个大小的配置参考tmp_table_size。

现分析下第二条SQL,,,子句id是一样的，那么执行顺序是从上至下，现查询pker表，索引为pkid,那么可以认为GROUP BY pker.pkid作用于上面的子句，rows为10，那么limit 10也是作用于此。

［优化索引、复合索引］

mysql> select count(*) from ng_game_pk WHERE status = 0 AND type = 4 AND uid = 1;
+----------+
| count(*) |
+----------+
|        0 |
+----------+
1 row in set (13.03 sec)
mysql> explain select count(*) from ng_game_pk WHERE status = 0 AND type = 4 AND uid = 1;
+----+-------------+------------+------+---------------+------+---------+-------+---------+-------------+
| id | select_type | table      | type | possible_keys | key  | key_len | ref   | rows    | Extra       |
+----+-------------+------------+------+---------------+------+---------+-------+---------+-------------+
|  1 | SIMPLE      | ng_game_pk | ref  | type          | type | 1       | const | 1729551 | Using where |
+----+-------------+------------+------+---------------+------+---------+-------+---------+-------------

这条语句用到了索引type，但是type取值范围很窄（1，2，3，4）其实这个索引没多大用处。
下面我们建立复合索引看效果如何，

mysql> alter table ng_game_pk add index idx_type_status_uid(type,status,uid);
Query OK, 5831851 rows affected (1 min 43.20 sec)
Records: 5831851  Duplicates: 0  Warnings: 0
mysql> explain select count(*) from ng_game_pk WHERE status = 0 AND type = 4 AND uid = 1;
+----+-------------+------------+------+--------------------------+---------------------+---------+-------------------+------+-------------+
| id | select_type | table      | type | possible_keys            | key                 | key_len | ref               | rows | Extra       |
+----+-------------+------------+------+--------------------------+---------------------+---------+-------------------+------+-------------+
|  1 | SIMPLE      | ng_game_pk | ref  | type,idx_type_status_uid | idx_type_status_uid | 6       | const,const,const |    1 | Using index |
+----+-------------+------------+------+--------------------------+---------------------+---------+-------------------+------+-------------+
1 row in set (0.11 sec)
mysql> select count(*) from ng_game_pk WHERE status = 0 AND type = 4 AND uid = 1;
+----------+
| count(*) |
+----------+
|        0 |
+----------+
1 row in set (0.00 sec)

看来有很好的效果了。对SQL稍作修改，

mysql> select sql_no_cache count(*) from ng_game_pk WHERE type = 4 and status>0 AND uid = 1;
+----------+
| count(*) |
+----------+
|     2649 |
+----------+
1 row in set (0.40 sec)

性能又下来了，这是因为B-Tree算法的原因，存储引擎将不能优化任何在第一个条件范围右边的列。那么就是(type,status,,,)后面的索引失效。

那么调整下，，，

mysql> drop index idx_type_status_uid on ng_game_pk;
mysql> alter table ng_game_pk add index idx_type_uid_status (type,uid,status);

结果性能又提升上来了。

－－－－－－－－－－－－－－－－－－－－－－－－－－－EOF－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－

参考：

＃《高性能MySql （第二版）》

＃《构建高性能web站点》

＃ http://www.cnitblog.com/aliyiyi08/archive/2008/09/09/48878.html

# http://www.perfgeeks.com/?p=460